Intrascene dynamic range refers to the range of incident light that can be accommodated by an image sensor in a single frame of pixel data. Examples of scenes that generate high dynamic range incident signals include an indoor room with outdoor window, an outdoor scene with mixed shadows and bright sunshine, night-time scenes combining artificial lighting and shadows and, in an automotive context, an auto entering or about to leave a tunnel or shadowed area on a bright day.
Dynamic range is measured as the ratio of the maximum signal that can be meaningfully imaged by a pixel of the imager to its noise level in the absence of light. Typical CMOS active pixel sensors (and charge coupled device (CCD)) sensors have a dynamic range from 60 to 75 dB. This corresponds to light intensity ratios of 1000:1 to about 5000:1. Noise in image sensors, including CMOS active pixel image sensors, is typically between 10 and 50 e-rms. The maximum signal accommodated is approximately 30,000 to 60,000 electrons. The maximum signal is often determined by the charge-handling capacity of the pixel or readout signal chain. Smaller pixels typically have smaller charge handling capacity.
Typical scenes imaged by cameras have lighting levels that generate signals on the order of 10–1,000 electrons under low light (1–100 lux), 1000–10,000 electrons under indoor light conditions (100–1000 lux), and 10,000->1,000,000 electrons (1000–100,000 lux) under outdoor conditions. To accommodate lighting changes from scene to scene, the so-called interscene dynamic range, an electronic shutter is used to change the integration time of all pixels in the arrays from frame to frame.
To cover a single scene that might involve indoor lighting (100 lux) and outdoor lighting (50,000 lux), the required intrascene dynamic range is of the order of 5,000:1 (assuming 10 lux of equivalent noise) corresponding to 74 dB. In digital bits, this requires 13–14 bits of resolution. However, most CMOS active pixel sensors have only 10 bits of output and 8 bits of resolution typically delivered to the user in most image formats such as JPEG. Companding of the data is often used to go from 10–12 bits to 8 bits. One type of companding is gamma correction, where roughly the square root of the signal is generated.
In order to accommodate high intrascene dynamic range, several different approaches have been proposed in the past. A common denominator of most approaches is performing signal companding within the pixel by having either a total conversion to a log scale (so-called logarithmic pixel) or a mixed linear and logarithmic response in the pixel. One example of a mixed linear and logarithmic circuit can be found in co-pending and commonly assigned patent application Ser. No. 10/226,127, filed Aug. 23, 2002, titled “A Wide Dynamic Range Linear-And-Log Active Pixel,” the disclosure of which is incorporated by reference herein.
These prior approaches have several major drawbacks. First, the knee point in a linear-to-log transition is difficult to control, leading to fixed pattern noise in the output image. Second, under low light, the log portion of the circuit is slow to respond, leading to lag. Third, a logarithmic representation of the signal in the voltage domain (or charge domain) means that small variations in signal due to fixed pattern noise leads to large variations in the represented signal.
Linear approaches have also been described where the integration time is varied during a frame to generate several different signals. This approach has architectural problems if the pixel is read out at different points in time since data must be stored in an on-board memory before the signals can be fused together. Another approach is to integrate two different signals in the pixel, one with low gain and one with high gain. However, the low gain portion of the pixel often presents color separation processing problems.